So, if you've been following the space industry really through, or even if you haven't, throughout the whole space age, one of the biggest challenges has been launch. How do we get things into space? And thanks to companies like SpaceX and Blue Origin and Firefly and Rocket Lab and others, the problem of access to space is starting to be more solved and it's more easily accessible, more access to space, more frequent launches, and more reliable access. So now we're looking ahead to what do we build in space now that we can have that reliable ready access to space and what kind of new capabilities can we build? Throughout most of the space age, the space economy has largely been driven by two mission applications, which is remote sensing, taking pictures or doing other types of data from space of Earth, and communications, which is moving data around from one place on Earth to another place on Earth. So now as we have new capabilities unlocked in space and the ability to build bigger, we're starting to see new missions that we can do and that growth in the space economy is going to come from new places beyond some of these initial applications. My name is Joe Landon. I'm the president and co-founder of Rendezvous Robotics. And at Rendezvous Robotics we are building spacecraft differently. Instead of building spacecraft in a factory and then folding them up to fit into a rocket for launch and then unfolding them once they get into space, we build modular spacecraft that flat pack into a rocket and then open up and autonomously self-assemble in space. So we can build megawatt-scale solar arrays and football-field-size antenna apertures to enable these new applications for communications, remote sensing, and beyond. And we're lucky today to have two practitioners who are really building the next wave of the space economy with us today. Philip Johnson, the CEO of StarCloud, and Will Marshall, the CEO of Planet. Philip, you want to go ahead and do a quick introduction to yourself and StarCloud?
Yeah, thanks so much for that intro, Joe. So I'm Philip Johnston. I'm the co-founder and CEO of StarCloud and we're building data centers in space in order to be able to take advantage of the abundant energy in space and the ability to scale almost indefinitely. So we're very quickly running up on energy constraints to build new data centers terrestrially and we think that can be solved by building large structures and large data centers in space. We just launched our first spacecraft StarCloud 1 in November that has the first Nvidia H100 on board. It's about 100 times more powerful GPU compute than it's been in space before. And with that, we've been training LLMs and running inference on various types of space data. Yeah, that's StarCloud.
Yeah, I'm Will Marshall, CEO of Planet. At Planet, we do Earth imaging. So we have a couple of hundred satellites in space imaging the entire Earth land mass every day. And Joe, just to add, the developments in space are in a renaissance right now. We've had lowering launch costs as you mentioned which have really helped galvanize the industry. But I would say the bigger thing that has happened, which is less publicly understood, is that satellite costs have come down, or increasing cost performance, about a thousand x over the last 10 years. And you say what, a thousand x? Actually a thousand x. So what's happened is we have taken satellites the size of this church or something and made them much more miniaturized and that's because we've taken modern electronics and stuffed them into smaller spacecraft that can do more and that thousand x is really what is driving the change in the number of satellites in space. Not the rocket cost. The rocket cost coming down 5x over the last 10 years has helped. But the capability per kilogram that we put into space has changed about a thousand x and that's been why we have suddenly Starlinks and Planet and other constant large fleets of satellites because the cost of the satellites has actually gone down considerably, orders of magnitude. And so I liken it to the mainframe to desktop computer revolution in IT. We took, you know, back in the mainframe days, every country maybe had one or two huge computers that cost a billion dollars and only used for specialized purposes. That was the era for space up until 10 or 20 years ago. We're opening up and now with desktops, there's a lot more applications, right, that suddenly become possible because of that lower cost point and one of them is compute in space. We might see space-based solar power, other things like that suddenly become viable because of the lower launch cost and the lower satellite costs.
Right. No, very good point Will and I want to dig into that but first I want to ask the audience to start thinking of your questions. We are going to do some audience Q&A toward the end of the session. So Will, on this excellent point, how spacecraft used to be very large, size of school buses, you know, hundreds of millions if not billions of dollars per, which means we had to design them to be very redundant and very robust because if you're going to spend that much on a satellite it better last a long time, right? So now we're starting to see, or we have seen over really the last 10 years, Planet was one of the pioneers in this, building smaller low-cost spacecraft. They don't need to last as long and they can have a more rapid update cycle. But now we're also seeing folks almost go the other way where what can we build very large platforms to do some of the things you mentioned like a solar power station or a data center. So Philip, tell us a little bit like what will your data centers look like? I think they'll be quite large, right?
Yeah. So the initial variants will be much smaller, so things that can fit through the Starship. Starship only has for the next sort of four to five years this very small slit to get your spacecraft out. So we'll be building for that form factor and we'll just launch many of those. Over time they will develop a clam shell door and we can deploy the full Starship payload bay. We're expecting to deploy about 10 megawatts per full Starship payload bay of compute, solar panels, and radiators and that will deploy into a very large multi-football-field-size surface area of solar panels and radiators. And then in the much longer term, sort of on a 10-year time frame, we're expecting to start docking these structures together. And the reason to dock them together in the instance of compute is number one, if you want to train a large model in space you need the whole thing to be internal, you can't do it with lasers. And number two, if you want to use the infrastructure for longer than the life of the chips. So the life of the chips might be five years, but your solar panels could last feasibly 20 years. Then you need to have the ability to dock things in and out of that. But yeah, in that instance, we're talking about a sort of 4 km by 4 km solar panel to produce 5 gigawatts of compute power and then about a 1 km by 4 km radiator in order to dissipate all of that heat. So pretty enormous structures that we're thinking about.
Got it. And I want to also give Will a chance to dig in a bit on Planet. So you mentioned that you image the whole Earth every day.
Yeah, absolutely. So back to the data question. Planet, we produce 4 million 47-megapixel images per day of the whole land mass of the Earth and some of the coastal water regions. And yeah, of course that's far too much for the individual human to look through all these images. And so we've been using AI on top of those data for now about a decade where we process that imagery so that, for example, we're working with the US Navy to look at the entire South China Sea for illicit vessel activities, challenging patterns that they then use to determine what operations to do. Or we're working with the Brazil Federal Police looking at all of the Brazilian Amazon, finding every day if there's illegal roads and buildings and then signs of illegal narcotics or illegal deforestation. And all of that is using AI. AI is enabling us to do scaled processing to answer questions about large parts of the Earth.
Now we're starting to have a lot of data so that we are starting to do some of that processing on the edge. So we've launched Nvidia chips, the Jetson Orin platform, on our latest satellites so that we can do some of the processing at the edge. But I think where things are going is really exciting. Firstly at the data level. So even when we bring the data to the ground, all the large language models that you use today are basically trained on the text of the internet and it's amazing what you can do with that, right? I mean, you can ask it questions about the knowledge of humanity. You can ask it to review a book. You could do some cool stuff, but imagine if you could ask it questions about the physical environment. Imagine the power of being able to ask Gemini or Claude or ChatGPT, pick your favorite LLM, if it could answer questions about the real physical planet and then you could answer questions about agriculture or insurance or disaster response or whatever going on around the world. You suddenly open up and unlock a whole, and that's where it's going because AI models are going to multimodality, that means not just text but images and videos and things, and satellite data has got too much data. It's got all this in principle value, but you need all the AI to pull it out.
So AI is touching every discipline out there. Medicine, weather modeling, all these things. But actually space and AI are getting married, I like to say, because they're so important for each other. AI models are all trying to do things about the real world. You see recently Demis, Dario, Fei-Fei, all these people all talking about how the AI models need to go into the real world. And for real world models, you need real world data. And that's where space comes in. We've got tons of that. In fact, Planet's probably got the most ideal set. Of course, I'm biased, but I really think that's true. But then, as I said, satellite data needs AI to extract out that value. And so, it's a match made in heaven. We're going to see AI and space get married. I like to think of Planet as the first truly space and AI company, meaning we are truly doing model development in AI and real hardware in space. And lots of people are playing at the edges of that, but we're really in the center of AI and space and I think we're going to see a number of companies become space and AI companies in the future.
Yeah, I think that's an excellent point. And if you think about why space and why we send things into space from the very beginning, it was because of this really the ultimate vantage point and the ultimate high ground. So the best place to gather the data about the whole Earth is from that vantage point where you could see the whole Earth and that's really what's driving a lot of this. So it's really interesting to hear how these AI models are starting to input the physical world and
I call it planetary intelligence because
instead of just large language models or artificial general intelligence
What about planetary intelligence? You know, I often thought when AI wakes up the first thing it's going to want is all the satellite data. Why? Because of course you first want a big picture perspective of what's going on.
What's the best route to that? All the satellites have their best picture of what's going on. So, you know, we'll see where the Borg wakes up or whatever is going to happen next. But it's an exciting time and space has a real meaningful role. But I think it's going to evolve this intelligence, artificial intelligence, towards a planetary intelligence where the sensors and the compute are going to be together. Now, the biology has thought of these things millions of years ago. You will notice that our eyes and ears are right next to our brain. There's a really good reason for that. You want to process all that data right there. You can't store it all. It's far too much information to store. So, you want to process it right at the edge there. Biology already figured this out. So, as we move towards planetary intelligence, we are going to want the compute and the imagery and the sensing systems right there together in space.
Yeah. So I want to push on that a little bit. So why not, we collect all this data in space, visual data, infrared data, radar data, all this data. Why not just send it down to Earth where we've got great computers, lots of resources, and process the data on Earth? I understand that like having it close to the processing, but you know, we can just send it down to Earth. Why not just do that on Earth?
Yeah. The main reason is because you have very low bandwidth or you're constrained in the amount of data that you can send from space to Earth because most of the time you're only above a ground station for 5 minutes of every 90-minute orbit. And when you are above a ground station, you're using very slow RF communication.
So it's just a real challenge to just send the radio signals from space, a satellite in space which is moving usually quite fast.
Satellites are collecting gobs of data and they can't get it all down. The bottleneck is RF.
Right. So there's a communications bandwidth constraint.
I mean that's the primary reason why we can't send all the data down. And so as Will mentioned, if we can process that data at the edge close to the source, we can draw insights before we just send down the insight. And the insight might be, in Will's case, there is a vessel in this location. That's like a kilobyte of data rather than 100 gigabytes of data that you might have had for the imagery.
That's right. I mean a typical one of our images, it might be hundreds of megabytes and as Philip is saying you can just do processing on the edge and send down the lat/long of a ship and that might take seconds. So I'd say the other piece of it is speed. It's sort of related, the bottleneck is speed. So people want answers immediately. I mean take the LA fires. We were helping out with that as soon as we could. We took a picture a couple of hours after the fires started. It took us a couple of hours more to do the AI processing, but then we got building by building damage reports out to the first responders, the firemen and the medical relief operators, FEMA and the like. But imagine if you could get that down to minutes. That could, you know, hours means lives in those situations. So getting that sort of AI processing back down as soon as possible really matters. And to do the processing on the edge is the way to do that. Otherwise, you have to downlink these gobs of data before you can then do it. So it's speed as well.
So that makes perfect sense. But I wonder, you know, Philip, if you could talk about, you know, there's a lot of AI processing, a lot of compute that happens on Earth. Does it, when does it make sense and will it make sense to send some of that data that we today process on Earth and do that work in space instead?
Yeah, that's a great question. So there will be a turning point when the launch cost gets below a certain point and we think that's around $500 per kilo. So about 10 times cheaper than today. There will be a turning point where it is cheaper in terms of kilowatt hours including the launch cost to do data processing in space. And I think that's likely we're going to see in sort of a 5-year time frame, volume productions of Starship, the new SpaceX launch vehicle. And then on a sort of 10-year time frame, it would not surprise me if we see most new data capacity being built in space.
And in order to do that, one of the constraints is how much power you can generate in space. That processing also generates a lot of heat, which is harder to get rid of in space in some cases than it is on Earth. There's certainly trade-offs there. So the ability to really build infrastructure in space is going to become a capability that we're going to need.
